So, when playing around with setpoints, I find that my error integral gets wound up to a large value in that it dominates everything else. While working out the problem, I have been running on PD mode (Setting K_i to zero).
Trying to find a better solution I hit up Wikipedia...